You know, gcc uses multiple passes, too, internally. If gcc were to use latex's model, you'd have to run it multiple times to get a working executable. Mind you, it would produce an executable every time, but those first ones would just crash. Latex can produce some nice looking documents, but as a technology, it's shoe strings and bubble gum.
The only problem is that the cross-references may be out of sync. This is usually not a serious barrier to readability.
Another gcc analogy is a dependence on a third-party library whose API keeps changing. Then you'd have to run gcc once to regenerate the library, and once on the client using the library.
Peeling that back a level, the analogy is between files like libfoo.a and foo_defs.h (for gcc) and files like section1.aux and document.bbl (for latex).
LaTeX will warn you when the references are out of sync, and there is no reason why it can't make a second (or third) pass internally without having to call (pdf)latex manually again and throwing .aux .log .other_random_stuff files everywhere.
So whip up a shell script and call it "better_latex" mark it +x and call the real exe multiple times. Now it's an improved version that does it internally, just like gcc
I just use LaTeX-Mk or a similar package, which takes care of multiple runs of LaTeX fairly well.
It could be done another way, but in the era that TeX was developed, it would have required more memory than was realistically available most of the time. I find that once I've run it twice, I can get away with single runs usually after that until I'm ready for the "definitive" version. Until then, I'm less concerned with whether the cross-reference is right, than whether there is a cross-reference in that location at all.